Method, device, medium and product for controlling water quantity in liquor brewing process

Abstract

A liquor brewing process water control method, device, medium and product. In the method, the history production data of the pit is cleaned and structured to generate a water control process knowledge base; the water content state characteristics and water control process parameters of the fermented grains are extracted, the correlation degree matrix is obtained by correlation analysis; the target process factor is screened out, and the water control intelligent control model is trained and generated; the fermented grains moisture detection value and the environmental temperature and humidity detection value are obtained and input into the water control intelligent control model to calculate the target water addition amount and the initial water addition rate; the water control instruction is generated according to the target water addition amount and the initial water addition rate and sent to the water addition device to execute the water addition operation, and the fermented grains temperature change data and the fermented grains moisture change data are collected in real time; the deviation value is calculated, and the current water addition rate of the water addition device is adjusted. The implementation of the scheme overcomes the defects of inaccurate and unstable water addition caused by manual experience.

Description

Technical Field

[0001] This application relates to the field of automated control technology in liquor brewing, specifically to a method, equipment, medium, and product for controlling the amount of water dispensed during liquor brewing. Background Technology

[0002] In the process of baijiu brewing, adding water is a crucial step in controlling the moisture content of the mash entering the fermentation tank. After the grains are steamed and the spirit is distilled, a certain amount of water needs to be added to meet the needs of microbial growth and metabolism, ensuring normal fermentation. In traditional production methods, operators manually control the amount and rate of water addition based on a given standard range for the initial moisture content, combined with their own experience and observation of the mash's condition. Through accumulated experience, they determine the appropriate water addition parameters for different production conditions.

[0003] However, the aforementioned water control method, which relies on manual experience and judgment, suffers from technical problems of insufficient accuracy and stability. Due to differences in the experience levels of different operators and the subjective nature of manual judgment regarding the state of the mash, the amount of water added to mash with the same moisture content varies significantly between different batches of production. This results in significant fluctuations in the moisture content of the mash entering the fermentation pit, affecting the stability of the fermentation process and the consistency of product quality. Summary of the Invention

[0004] To address the aforementioned technical problems, this application provides a method, equipment, medium, and product for controlling the amount of water used in the brewing process of baijiu (Chinese liquor).

[0005] The first aspect of this application provides a method for controlling the amount of water added during the brewing process of baijiu (Chinese liquor), which adopts the following technical solution: Collect historical production data of pits for a preset number of production cycles, clean and structure the historical production data of the pits, and generate a water quality process knowledge base. The moisture state characteristics of the mash and the parameters of the water-measuring process are extracted from the water-measuring process knowledge base. A correlation matrix is ​​obtained by performing a correlation analysis on the moisture state characteristics of the mash and the water-measuring process parameters. Target process factors are selected based on the correlation matrix, and a water intelligent control model is trained and generated based on the target process factors. The moisture content of the mash and the ambient temperature and humidity of the fermentation pit to be treated are obtained. The moisture content of the mash and the ambient temperature and humidity are input into the intelligent water addition control model to calculate the target water addition amount and the initial water addition rate. A water addition control command is generated based on the target water addition amount and the initial water addition rate and sent to the water addition equipment to execute the water addition operation. During the water addition operation, the temperature change data and moisture change data of the mash are collected in real time. Based on the temperature change data and moisture change data of the mash, the deviation value between the current process state and the preset process target is calculated, and the current water addition rate of the water addition equipment is adjusted according to the deviation value until the water addition operation is completed.

[0006] By adopting the above technical solutions, a knowledge base for water addition processes is constructed and an intelligent control model is trained, enabling accurate prediction of water addition volume and rate. This overcomes the defects of inaccurate and unstable water addition caused by human experience, effectively reduces moisture fluctuations between batches, and improves fermentation stability and consistency of liquor product quality.

[0007] Optionally, the correlation analysis of the moisture state characteristics of the mash and the water mixing process parameters to obtain a correlation matrix includes: The moisture state characteristics of the fermented mash and the water mixing process parameters are normalized to obtain a standardized feature dataset; Based on the standardized feature dataset, the Pearson correlation coefficient between the moisture state characteristics of each mash and the water mixing process parameters is calculated, and a correlation coefficient matrix is ​​generated. Obtain the number of samples in the standardized feature dataset and calculate the degree of freedom of the samples. Determine the significance probability value corresponding to each correlation coefficient based on the correlation coefficient matrix, the degree of freedom of the samples, and the preset probability distribution model. Select the correlation coefficients whose significance probability value is less than the preset significance level threshold as effective correlation coefficients. Extract the target mash moisture state feature identifier and the target water process parameter identifier corresponding to the effective correlation coefficient. The target mash moisture state feature identifiers are sorted according to a preset sorting rule and mapped sequentially to matrix row indices. The target water process parameter identifiers are sorted according to a preset sorting rule and mapped sequentially to matrix column indices. A two-dimensional matrix index structure is constructed based on the matrix row indices and the matrix column indices. Based on the target mash moisture state characteristic identifier and the target volume water process parameter identifier, the corresponding row index value and column index value are found in the two-dimensional matrix index structure. The filling coordinates are determined based on the row index value and column index value, and the effective correlation coefficients are filled to the corresponding index positions in the two-dimensional matrix index structure according to the filling coordinates to generate the correlation matrix.

[0008] By adopting the above technical solution, the effective correlation between the moisture characteristics of mash and process parameters can be accurately identified. By combining the sorting mapping to construct a structured correlation matrix, the scientific nature of feature selection and the standardization of matrix construction are improved, and the accuracy and reliability of model input are enhanced.

[0009] Optionally, the step of obtaining the moisture content of the mash and the ambient temperature and humidity of the fermentation pit to be treated, and inputting the moisture content of the mash and the ambient temperature and humidity into the intelligent water addition control model to calculate the target water addition amount and the initial water addition rate includes: The moisture content data of the mash in the fermentation pit to be treated is collected by a preset moisture sensor to obtain the moisture detection value of the mash. Temperature and humidity data of the environment where the pit to be treated is located are collected by a preset temperature and humidity sensor to obtain the environmental temperature and humidity detection values. The moisture content of the fermented mash and the ambient temperature and humidity are normalized to map them to a preset range, thereby generating normalized input data. The normalized input data is combined into an input feature vector according to a preset feature arrangement order, and the input feature vector is input into the intelligent control model for measuring water. The input feature vector is calculated by forward propagation using the intelligent water control model, and the target water addition amount and the initial water addition rate are output through the output layer of the intelligent water control model.

[0010] By adopting the above technical solution, the moisture content of the mash and the ambient temperature and humidity data are collected collaboratively, and normalized and vectorized to ensure the standardization and consistency of the input data. Combined with the forward propagation calculation of the intelligent model, the target water addition amount and the initial water addition rate are accurately output, which improves the accuracy and adaptability of the control parameter prediction.

[0011] Optionally, the step of calculating the deviation between the current process state and the preset process target based on the temperature change data and moisture change data of the mash, and adjusting the current water addition rate of the water addition equipment according to the deviation value, includes: The temperature change data and moisture change data of the mash are arranged in time series to construct temperature change curves and moisture change curves of the mash. The current temperature rise rate is calculated based on the temperature change curve of the mash, and the current moisture absorption rate is calculated based on the moisture change curve of the mash. The standard temperature rise rate and standard moisture absorption rate are extracted from the preset process target database. The temperature deviation value is calculated by the difference between the current temperature rise rate and the standard temperature rise rate. The moisture deviation value is calculated by the difference between the current moisture absorption rate and the standard moisture absorption rate. The temperature deviation value and the moisture deviation value are input into a preset feedback control algorithm. The feedback control algorithm performs calculations based on a preset proportional coefficient, integral time, and derivative time to generate a water addition rate adjustment amount. The adjusted water filling rate is superimposed with the current water filling rate to obtain the adjusted water filling rate. The adjusted water filling rate is then updated to the new current water filling rate, and a rate adjustment command is generated based on the adjusted water filling rate and sent to the water filling device.

[0012] By adopting the above technical solution, temperature and moisture change curves are constructed, deviation values ​​are calculated in real time, and a feedback control algorithm is introduced to dynamically adjust the water addition rate, thereby realizing closed-loop control of the irrigation process. This effectively improves the accuracy and response speed of water addition operations, ensures stable process conditions, and enhances the controllability and consistency of the fermentation process.

[0013] Optionally, the step of inputting the temperature deviation value and the moisture deviation value into a preset feedback control algorithm, and generating a water addition rate adjustment amount by the feedback control algorithm based on a preset proportional coefficient, integral time, and derivative time, includes: The temperature deviation value and the moisture deviation value are weighted and summed according to a preset weighting coefficient to obtain a comprehensive deviation value; The comprehensive deviation value is multiplied by the preset proportional coefficient to obtain the proportional adjustment component; Perform time integration on the comprehensive deviation value, and divide the integration result by the preset integration time to obtain the integral adjustment component; Perform time differentiation on the comprehensive deviation value, and multiply the result of the differentiation operation by the preset differentiation time to obtain the differential adjustment component; The proportional adjustment component, the integral adjustment component, and the derivative adjustment component are summed to obtain the preliminary adjustment amount; The initial adjustment amount is subjected to amplitude limiting processing to restrict the initial adjustment amount within a preset adjustment range, thereby generating the water addition rate adjustment amount.

[0014] By adopting the above technical solution, the temperature and moisture deviations are weighted and integrated, and dynamic adjustment quantities are generated by combining proportional, integral, and differential operations. This enables precise closed-loop control of the water addition rate, effectively suppressing overshoot and oscillation, improving system response stability and control accuracy, and ensuring the continuity and consistency of the brewing process.

[0015] Optionally, the method further includes: The moisture content of the mash after water addition is measured using a pre-set moisture detection device, and the measured moisture value of the mash is obtained. Extract the preset moisture standard corresponding to the fermentation pit to be treated from the preset process standard database, calculate the difference between the measured moisture value of the mash entering the pit and the preset moisture standard, and obtain the moisture control deviation value. The prediction accuracy of the intelligent water control model is determined based on the magnitude and direction of the moisture control deviation value. The model weight adjustment coefficient and model bias adjustment coefficient are calculated based on the prediction accuracy to generate the model correction parameters. The measured values ​​of the mash moisture content, the measured values ​​of the ambient temperature and humidity, the target water addition amount, and the measured values ​​of the mash moisture content entering the fermentation tank are combined to form training sample data. The training sample data and the model correction parameters are then input into the intelligent water control model. The intelligent water control model adjusts the weight parameters and bias parameters within the model according to the model correction parameters, and updates the network parameters of the intelligent water control model through a preset backpropagation algorithm to obtain the target intelligent water control model.

[0016] By adopting the above technical solution, introducing a measured moisture feedback and model correction mechanism, and combining the backpropagation algorithm to dynamically optimize the model parameters, the intelligent water control model can achieve self-learning and self-adaptation, continuously improve the accuracy of water addition prediction, and enhance the system's adaptability and long-term stability to changing working conditions.

[0017] Optionally, the step of determining the prediction accuracy of the intelligent water control model based on the magnitude and direction of the moisture control deviation value, calculating the model weight adjustment coefficient and model bias adjustment coefficient based on the prediction accuracy, and generating the model correction parameters includes: The absolute value of the moisture control deviation value is calculated to obtain the deviation amplitude value. The deviation amplitude value is compared with a preset set of deviation level thresholds. Based on the comparison result, the deviation level corresponding to the moisture control deviation value is determined. The direction of water addition control is determined based on the positive or negative sign of the moisture control deviation value. When the moisture control deviation value is positive, it is determined that the water addition is insufficient; when the moisture control deviation value is negative, it is determined that the water addition is excessive. Based on the deviation level, the corresponding prediction accuracy score is retrieved from the preset accuracy mapping table, and the learning rate adjustment factor is calculated based on the prediction accuracy score. The learning rate adjustment factor is multiplied by the preset benchmark learning rate to obtain the current learning rate. The model weight adjustment coefficient and the model bias adjustment coefficient are calculated based on the current learning rate and the water addition control direction. The model weight adjustment coefficient, the model bias adjustment coefficient, and the current learning rate are encapsulated into a parameter data structure to generate the model correction parameters.

[0018] By adopting the above technical solution, model correction parameters are dynamically generated. Combined with the adaptive learning rate adjustment mechanism, the model weights and biases are accurately updated, which enhances the self-optimization capability of the intelligent control model and improves the stability and prediction accuracy of long-term operation.

[0019] A second aspect of this application provides an electronic device including a processor, a memory, a user interface, and a network interface, wherein the memory is used to store instructions, the user interface and the network interface are both used to communicate with other devices, and the processor is used to execute the instructions stored in the memory to cause the electronic device to perform the method as described in any of the foregoing.

[0020] A third aspect of this application provides a computer-readable storage medium storing instructions that, when executed, perform the method described in any of the preceding descriptions.

[0021] A fourth aspect of this application provides a computer program product that, when run on an electronic device, causes the electronic device to perform the method as described in any of the preceding claims.

[0022] In summary, one or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages: By constructing a comprehensive water addition knowledge base and intelligent control model, combined with multi-source sensor data and a closed-loop feedback mechanism, precise and intelligent control of the water addition process in baijiu brewing has been achieved. Key process factors are screened through correlation analysis to improve the scientific rigor of the model input; real-time deviation feedback and adaptive learning mechanisms are introduced to dynamically optimize water addition parameters, effectively reducing batch-to-batch fluctuations and improving fermentation stability and product quality consistency. This is significantly superior to traditional extensive management methods that rely on manual experience, demonstrating significant industrial application value. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the system architecture of an embodiment of a water quantity control method in the brewing process of baijiu (Chinese liquor) according to this application; Figure 2 This is a schematic flowchart of a method for controlling the amount of water during the brewing process of baijiu (Chinese liquor) as disclosed in an embodiment of this application. Figure 3 This is a schematic diagram of the structure of an electronic device disclosed in an embodiment of this application.

[0024] Explanation of reference numerals in the attached figures: 100, System architecture; 101, First terminal device; 102, Second terminal device; 103, Third terminal device; 104, Network; 105, Server; 301, Processor; 302, Communication bus; 303, User interface; 304, Network interface; 305, Memory. Detailed Implementation

[0025] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0026] In the description of the embodiments in this application, terms such as "for example" or "by way of example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "for example" or "by way of example" in the embodiments of this application should not be construed as being better or more advantageous than other embodiments or design solutions. Rather, the use of terms such as "for example" or "by way of example" is intended to present the relevant concepts in a specific manner.

[0027] In the description of the embodiments of this application, the term "multiple" means two or more. For example, "multiple systems" refers to two or more systems, and "multiple screen terminals" refers to two or more screen terminals. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. The terms "including," "comprising," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized.

[0028] like Figure 1 As shown, system architecture 100 may include terminal devices 101, 102, and 103, a network 104, and a server 105. Network 104 serves as the medium for providing communication links between terminal devices 101, 102, and 103 and server 105. Network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0029] Users can use terminal devices 101, 102, and 103 to interact with server 105 via network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, and 103, such as model training applications, video recognition applications, web browser applications, social platform software, etc.

[0030] Terminal devices 101, 102, and 103 can be either hardware or software. When terminal devices 101, 102, and 103 are hardware, they can be various electronic devices with displays, including but not limited to smartphones, tablets, e-book readers, MP3 (Moving Picture Experts Group Audio Layer III) players, MP4 (Moving Picture Experts Group Audio Layer IV) players, laptops, and desktop computers, etc. When terminal devices 101, 102, and 103 are software, they can be installed in the aforementioned electronic devices. They can be implemented as multiple software programs or software modules (e.g., multiple software programs or software modules used to provide distributed services) or as a single software program or software module. No specific limitations are imposed here.

[0031] This embodiment discloses a method for controlling the amount of water added during the brewing process of baijiu (Chinese liquor). Figure 2 This is a schematic flowchart of a method for controlling the amount of water used in the brewing process of baijiu (Chinese liquor) disclosed in an embodiment of this application. Figure 2 As shown, the method includes the following steps: S201. Collect historical production data of pits for a preset number of production cycles, clean and structure the historical production data of the pits, and generate a water-processing knowledge base. In step S201, the core task is to construct a high-quality, digital knowledge base for the water addition process, laying a solid data foundation for subsequent intelligent model training. This process begins with the comprehensive collection of historical production data. For example, systematically collect fermentation data from the past 1 to 3 years, covering multiple complete production cycles (e.g., the 4 cycles mentioned in the technical briefing) and different seasons, to ensure the diversity and representativeness of the data sample. The data source can be automated, such as through integration with the distillery's production execution system or data acquisition and monitoring control system; or it can be semi-automated, that is, digitizing historical paper production records (e.g., the water addition data collection form provided in the technical briefing). The collected data fields may include, but are not limited to: production batch number, shift identification (e.g., shift 3, shift 4), fermentation pit number, stilling batch, moisture content of mash before water addition (%), water addition volume (L), water addition temperature (°C), water addition method (e.g., manual sprinkling, machine addition), moisture content of mash after water addition (%), final moisture content entering the fermentation pit (%), and the ambient temperature and humidity of the workshop during the same period. After data collection, rigorous cleaning and structuring processes will be performed. A specific cleaning scheme involves first identifying and removing obvious logical errors or outliers using pre-defined business rules (e.g., moisture content should be between 30% and 70%, and water temperature should be between 40℃ and 100℃). For missing data items, mean, median, or time-series-based interpolation methods can be used to fill in the missing data based on the context of the batch. Subsequently, the cleaned data will be structured, for example, into a two-dimensional data table, where each row represents an independent production record (e.g., a single batch of water measurement), and each column corresponds to a feature dimension, as shown in Table 1 (a sample structure). Ultimately, this dataset, containing a vast amount of historical production cases and having undergone purification and standardization, constitutes the water measurement process knowledge base referred to in this application. It can be stored in a relational database (such as MySQL) or managed in file format (such as CSV or Parquet), providing a standardized, high-quality data source for subsequent correlation analysis and model building.

[0032] S202. Extract the moisture state characteristics of the mash and the water mixing process parameters from the water mixing process knowledge base, and perform a correlation analysis on the moisture state characteristics of the mash and the water mixing process parameters to obtain a correlation matrix. After constructing the knowledge base for the water addition process, the system executes step S202. Its core objective is to quantify and filter out the key factors that have the most significant impact on the water addition process from massive historical data using scientific statistical analysis methods. Specifically, the system first extracts two types of key variables from the knowledge base: the first type describes the moisture state characteristics of the mash under initial brewing conditions, such as, but not limited to, the moisture content of the mash before adding water, ambient temperature, and ambient humidity; the second type describes the water addition process parameters, such as the amount of water added, the temperature of the water addition, and the final moisture content entering the fermentation tank. To eliminate the interference of different physical units (such as percentage of moisture and °C of temperature) on the correlation calculation, the system performs data normalization on these two types of data. One specific implementation is to use the Z-score normalization method to convert the data of each feature into a standard normal distribution with a mean of 0 and a standard deviation of 1, generating a standardized feature dataset. As another optional solution, max-min normalization can also be used to linearly map the data to the [0, 1] interval. Based on this standardized dataset, the system calculates the Pearson correlation coefficient between the moisture state characteristics of each mash and the process parameters of each batch of water, generating a preliminary correlation coefficient matrix. Each value in the matrix (ranging from -1 to +1) initially quantifies the strength and direction of the linear correlation between a pair of variables. However, to ensure that the discovered associations are statistically significant and not random coincidences, the system further performs a significance test. Specifically, the system obtains the sample size n of the standardized feature dataset and calculates the sample degrees of freedom (usually n-2). Then, based on each coefficient value in the correlation coefficient matrix, the sample degrees of freedom, and a preset probability distribution model (such as a t-distribution), it calculates the significance probability value (P-value) corresponding to each correlation coefficient. The system compares this P-value with a preset significance level threshold (e.g., α=0.05). Only when the P-value is less than this threshold is the correlation coefficient considered statistically significant and selected as a valid correlation coefficient. Simultaneously, the system extracts the target mash moisture state characteristic identifier and the target batch water process parameter identifier corresponding to this valid correlation coefficient. Finally, to generate a structured and easily interpretable final output, the system sorts the selected target mash moisture state feature identifiers according to a preset sorting rule (e.g., alphabetical order of feature names) and maps them sequentially to row indices of a matrix. The target water-refining process parameter identifiers are also sorted according to the same rule and mapped to column indices, thus constructing a two-dimensional matrix index structure. Subsequently, based on each pair of target feature and parameter identifiers, the system searches for the corresponding row and column index values ​​in the constructed two-dimensional matrix index structure and fills the unique coordinate positions determined by those row and column index values ​​with their corresponding valid correlation values. This ultimately generates a clear and quantifiable correlation matrix, which intuitively reveals the key driving factors that have a decisive impact on the water-refining process.

[0033] Optionally, the step of performing correlation analysis on the moisture state characteristics of the fermented mash and the water mixing process parameters to obtain a correlation matrix includes: performing data normalization processing on the moisture state characteristics of the fermented mash and the water mixing process parameters to obtain a standardized feature dataset; based on the standardized feature dataset, calculating the Pearson correlation coefficient between each moisture state characteristic of the fermented mash and each water mixing process parameter to generate a correlation coefficient matrix; obtaining the sample size of the standardized feature dataset and calculating the sample degrees of freedom; determining the significance probability value corresponding to each correlation coefficient according to each correlation coefficient in the correlation coefficient matrix, the sample degrees of freedom, and a preset probability distribution model; selecting correlation coefficients with significance probability values ​​less than a preset significance level threshold as effective correlation coefficients, and extracting the relevant correlation coefficients. The effective correlation coefficient corresponds to the target mash moisture state feature identifier and the target volumetric water process parameter identifier; the target mash moisture state feature identifier is sorted according to a preset sorting rule and mapped sequentially to a matrix row index, and the target volumetric water process parameter identifier is sorted according to a preset sorting rule and mapped sequentially to a matrix column index, and a two-dimensional matrix index structure is constructed based on the matrix row index and the matrix column index; the corresponding row index value and column index value are found in the two-dimensional matrix index structure according to the target mash moisture state feature identifier and the target volumetric water process parameter identifier, the filling coordinates are determined based on the row index value and the column index value, and the effective correlation coefficient is filled to the corresponding index position of the two-dimensional matrix index structure according to the filling coordinates to generate the correlation degree matrix.

[0034] First, the system performs data normalization on the moisture state features of the mash (such as moisture content before water addition and ambient temperature) and water addition process parameters (such as water addition amount and water temperature) extracted from the water addition process knowledge base to generate a standardized feature dataset. The purpose of this step is to eliminate the unequal weighting of different variables when calculating correlations due to differences in their physical units and numerical ranges (dimensions). One specific implementation is to use the Z-score normalization method, which transforms each column of feature data using the formula z = (x - μ) / σ (where x is the original value, μ is the average value of the feature column, and σ is the standard deviation), ensuring that the transformed data follows a standard normal distribution with a mean of 0 and a standard deviation of 1. Alternatively, a max-min normalization method can be used, using the formula x' = (x - min) / (max - min), where x is a specific original data point to be processed, and min and max represent the minimum and maximum values ​​in the entire feature dataset (i.e., the data column) to which the data point belongs, linearly scaling the data to a preset interval, such as [0, 1]. Regardless of the method used, the resulting standardized feature dataset lays the foundation for subsequent fair comparisons and accurate calculations.

[0035] Furthermore, based on this standardized feature dataset, the system calculates the Pearson correlation coefficients between each mash moisture state characteristic and each water addition process parameter, thereby generating a preliminary correlation coefficient matrix. The Pearson correlation coefficient is a statistic used to measure the strength and direction of the linear relationship between two continuous variables, with a value range of [-1, 1]. The system calculates the correlation coefficients between all possible feature-parameter pairs. For example, it calculates the correlation coefficient between the moisture content of the mash before water addition and the amount of water added, as well as the correlation coefficient between ambient temperature and the amount of water added. The calculation results are organized into a matrix, where the rows of the matrix represent the mash moisture state characteristics, the columns represent the water addition process parameters, and the element value at position (i, j) in the matrix is ​​the correlation coefficient r between the i-th feature and the j-th parameter.

[0036] Furthermore, to ensure that the calculated correlations are not due to random sampling errors, the system performs a significance test to filter out truly statistically significant associations. This process first involves obtaining the sample size n of the standardized feature dataset and calculating the sample degrees of freedom (df). For a Pearson correlation test of two variables, the degrees of freedom are typically n-2. Then, based on each correlation coefficient value r in the correlation coefficient matrix, the calculated degrees of freedom df, and a pre-defined probability distribution model (usually a t-distribution), the system performs a t-test using the formula t=r×sqrt(df / (1-r)). 2 The system calculates the t-statistic and further determines the significance probability value (P-value) corresponding to it. This P-value is compared to a preset significance level threshold (e.g., α = 0.05 or 0.01). Only when the P-value is less than this threshold is the null hypothesis rejected (i.e., no correlation is considered between the two variables), and the corresponding correlation coefficient is considered a valid correlation coefficient. While selecting valid correlation coefficients, the system also extracts and records the target mash moisture state feature identifier (e.g., Feature_InitialMoisture) and target volumetric water process parameter identifier (e.g., Param_WaterVolume) corresponding to these coefficients.

[0037] Furthermore, after obtaining all valid correlation coefficients and their corresponding feature and parameter identifiers, the system constructs a standardized two-dimensional matrix index structure to facilitate subsequent model invocation and manual interpretation. Specifically, the system first sorts all selected target mash moisture state feature identifiers according to a preset sorting rule (e.g., arranged in ascending alphabetical order), and then maps the sorted sequence to the row indices of the matrix (e.g., the first identifier corresponds to row index 0, the second to row index 1, and so on). Similarly, the system also sorts all selected target mash moisture process parameter identifiers according to the same or different sorting rules and maps them to the column indices of the matrix. In this way, a clear two-dimensional matrix logical framework is constructed, with each row and column having a clear business meaning.

[0038] Furthermore, based on the index structure constructed in the previous step, the system fills in the effective correlation coefficients to generate the final correlation matrix. For each selected effective correlation coefficient, the system searches for its corresponding row and column index values ​​in the index mapping relationship constructed in the previous step, according to the target mash moisture state characteristic identifier and the target volumetric water process parameter identifier. These two index values ​​together determine the unique filling coordinate of the effective correlation coefficient in the matrix. The system fills the value of the effective correlation coefficient into the index position corresponding to this coordinate. After traversing all effective correlation coefficients and completing the filling, a structured, quantified, and statistically significant correlation matrix is ​​officially generated. This matrix intuitively shows which mash states have a significant impact on which process parameters, as well as the degree and direction of the impact.

[0039] S203. Select target process factors based on the correlation matrix, and train and generate a water intelligent control model based on the target process factors. After obtaining and analyzing the correlation matrix, the system further performs the screening of target process factors and trains a water-quantity intelligent control model based on this. Specifically, the system first screens out the target process factors that have the most significant impact on the water-quantity process based on the correlation matrix. One specific implementation is to set a threshold for the absolute value of the correlation coefficient, such as 0.6. The system will traverse the correlation matrix and identify all mash moisture state features whose absolute correlation coefficient with key output parameters (such as the final mash moisture content or the actual amount of water added) is greater than this threshold. These screened features with strong correlation (such as mash moisture content before water addition, ambient temperature, etc.) constitute the set of target process factors. This set significantly reduces the feature dimension of subsequent model training and eliminates the interference of noisy data. As another optional screening scheme, the system can also directly select the top K features with the highest absolute value of their correlation coefficient with a certain key output parameter as target process factors, where K is a positive integer that can be preset according to actual needs. After screening, the system uses these target process factors as input features and the corresponding key process parameters (e.g., water addition) as prediction labels to train a water intelligent control model. The model selection is diverse; for example, a gradient boosting decision tree (GBDT) model based on ensemble learning can be used. During training, the system divides the historical dataset containing the target process factors and corresponding labels into training and validation sets. The model iteratively learns on the training set, continuously fitting the complex nonlinear relationship between input features and output labels. Its core optimization objective is to minimize the loss function, such as the mean squared error (MSE), calculated as follows: Where n represents the total number of training samples, yi represents the actual water addition value of the i-th sample, and ŷi represents the water addition value predicted by the model based on the input features of the i-th sample. The Σ symbol represents the summation of the squared errors of all samples. The system continuously adjusts the internal parameters of the model through optimization algorithms such as gradient descent to minimize the MSE value. When the model's performance on the validation set reaches a preset standard or no longer improves, the training process terminates, ultimately resulting in a smart water control model that can accurately predict the optimal water addition amount based on the real-time characteristics of the fermented mash (i.e., the target process factor).

[0040] S204. Obtain the moisture content of the mash and the ambient temperature and humidity of the fermentation pit to be treated, and input the moisture content of the mash and the ambient temperature and humidity into the intelligent water addition control model to calculate the target water addition amount and the initial water addition rate. When a specific fermentation pit enters the water measurement process, it enters the practical application stage. First, the system acquires key status data of the pit in real time through sensors deployed on-site. Specifically, accurate moisture content values ​​of the mash can be obtained through an online moisture meter installed above the mash conveyor belt, or by operators using a handheld moisture detection probe. Simultaneously, ambient temperature and humidity sensors installed in the brewing workshop near the pit collect current ambient temperature and humidity values. These values ​​are transmitted to the central control system via an industrial bus (such as RS485) or a wireless network (such as Wi-Fi). Subsequently, the system constructs a feature vector from these real-time acquired values ​​(e.g., mash moisture 55.2%, ambient temperature 28°C, ambient humidity 75%), conforming to the input format of the intelligent water measurement control model. This feature vector is then input into the model trained in the previous step for inference calculations. The model utilizes its internally learned complex mapping relationships to rapidly compute the input vector and output two core process control parameters: one is the target water volume (e.g., 150 liters) to guide the total water volume, and the other is the initial water addition rate (e.g., 25 liters per minute) to initiate the water addition process. These two calculation results will directly serve as the command basis for the next stage of automated control.

[0041] Optionally, the step of obtaining the moisture content of the mash and the ambient temperature and humidity values ​​in the fermentation pit to be treated, and inputting the moisture content of the mash and the ambient temperature and humidity values ​​into the intelligent water addition control model to calculate the target water addition amount and the initial water addition rate, includes: collecting the moisture content data of the mash in the fermentation pit to be treated through a preset moisture sensor to obtain the moisture content of the mash; collecting the temperature and humidity data of the environment where the fermentation pit is located through a preset temperature and humidity sensor to obtain the ambient temperature and humidity values; and performing the moisture content measurement on the mash. The moisture content of the mash and the ambient temperature and humidity are normalized to map the moisture content of the mash and the ambient temperature and humidity to a preset numerical range, generating normalized input data. The normalized input data is combined into an input feature vector according to a preset feature arrangement order, and the input feature vector is input to the intelligent water addition control model. The intelligent water addition control model performs forward propagation calculation on the input feature vector, and outputs the target water addition amount and the initial water addition rate through the output layer of the intelligent water addition control model.

[0042] As a preferred implementation of the foregoing embodiments, the steps of acquiring real-time data of the pit to be treated and calculating control parameters can be broken down into the following consecutive sub-steps to ensure the accuracy and reliability of the entire control process.

[0043] The first step is the acquisition of real-time status data. In this embodiment, the system accomplishes this task through pre-deployed automated sensing equipment. Specifically, a near-infrared online moisture detector can be installed above the conveyor belt from the cooking pot to the water measurement area. This detector can continuously and non-contactly scan the mash flowing below, measuring its moisture content in real time. The system collects continuous readings over a stable time period (e.g., 5 seconds) and calculates the average value, which is used as the final moisture content value of the mash, effectively avoiding random errors from single-point measurements. Simultaneously, one or more industrial-grade temperature and humidity sensors are pre-positioned near the fermentation pits, representing the local macroscopic environment of the workshop. These sensors continuously monitor the ambient temperature and relative humidity, transmitting real-time data to the control system as ambient temperature and humidity detection values.

[0044] Furthermore, the collected raw data is preprocessed to meet the input requirements of the intelligent water control model. Since the model uses normalized data during training, the same normalization process must be performed on newly entered real-time detection data. Specifically, the system uses the same normalization parameters as during training (e.g., the maximum and minimum values ​​of the moisture feature in the training dataset). Assuming the max-min normalization method is used, the system linearly maps the real-time mash moisture detection values ​​and environmental temperature and humidity detection values ​​to the preset numerical range [0, 1], generating normalized input data. After normalization, the system combines these values ​​into a structured input feature vector according to the preset feature arrangement order determined during model training, such as [normalized mash moisture value, normalized environmental temperature value, normalized environmental humidity value].

[0045] Furthermore, the system formally inputs the input feature vector generated in the previous step into the intelligent water control model that has already been loaded into memory. Upon receiving the vector, the model performs a complete forward propagation calculation. During this process, the input feature vector data is passed layer by layer between multiple hidden layers of the model, with each layer undergoing a complex nonlinear transformation through an activation function. Finally, the calculation result is passed to the model's output layer. This output layer is specially designed to contain two output neurons: one corresponding to the predicted target water addition amount, and the other corresponding to the predicted initial water addition rate. The model outputs the activation values ​​of these two neurons as the final calculation result, thus efficiently and accurately completing the entire process from real-time state perception to specific process parameter decision-making.

[0046] S205. Generate a water addition control command based on the target water addition amount and the initial water addition rate and send it to the water addition equipment to execute the water addition operation. During the water addition operation, collect real-time data on changes in mash temperature and moisture content. Specifically, after the intelligent water control model outputs the target water addition volume and initial water addition rate, the central control system immediately converts these two core parameters into specific, executable equipment control instructions. For example, the system generates an instruction set containing the target flow rate (i.e., the initial water addition rate) and the cumulative flow limit (i.e., the target water addition volume), and sends it to the field programmable logic controller (PLC) responsible for the water addition operation via industrial Ethernet or fieldbus protocols (such as Modbus TCP / IP). After receiving the instructions, the PLC precisely controls the water pump connected to the frequency converter and the electric regulating valve on the pipeline, stabilizing the initial water output rate at the value calculated by the model. At the same time, to achieve dynamic monitoring and closed-loop feedback of the process, the system synchronously activates the real-time data acquisition module at the moment the water addition operation starts. Specifically, an insertion-type temperature sensor (e.g., a Pt100 RTD) installed in the mixing area of ​​the mash and water begins to collect the real-time temperature of the mixed mash at a high frequency (e.g., once per second), forming a continuous temperature change curve. Meanwhile, an online moisture detector located above the conveyor belt downstream of the water addition point also operates continuously, measuring the moisture content of the mash after water addition and initial mixing in real time, generating another crucial data stream of moisture changes. These two sets of real-time collected data are continuously transmitted back to the central control system, providing immediate and accurate on-site status information for any subsequent dynamic adjustments, thus ensuring that the entire water addition process is not only automated but also intelligent and adaptive.

[0047] S206. Calculate the deviation between the current process state and the preset process target based on the temperature change data and moisture change data of the mash, and adjust the current water addition rate of the water addition equipment according to the deviation value until the water addition operation is completed.

[0048] During the water addition operation, the system continuously compares the collected data on changes in mash temperature and moisture content with a preset ideal process target curve stored in the system. This target curve defines the ideal temperature and moisture values ​​that the mash should have at different water addition stages. In each control cycle (e.g., every second), the control system calculates the deviation between the current actual measurement value and the corresponding point on the target curve. This deviation is then input into a control algorithm module, such as a classic proportional-integral-derivative (PID) controller. This controller calculates a precise adjustment amount in real time based on the magnitude, duration, and trend of the deviation. For example, if the rate of increase in mash moisture content is detected to be lower than the preset target, the PID controller outputs a positive adjustment command, moderately increasing the current water addition rate by increasing the operating frequency of the water pump inverter; conversely, if the temperature drops too quickly beyond a preset threshold, the water addition rate will be reduced. This cycle of data collection, comparison, calculation, and adjustment will continue uninterrupted until the cumulative water addition reaches the target water addition amount set in step S205, thereby ensuring that the entire water addition process is always carried out on the optimal process path.

[0049] Optionally, the step of calculating the deviation between the current process state and the preset process target based on the mash temperature change data and the mash moisture change data, and adjusting the current water addition rate of the water addition equipment according to the deviation value, includes: arranging the mash temperature change data and the mash moisture change data into a time series to construct a mash temperature change curve and a mash moisture change curve; calculating the temperature rise rate at the current moment based on the mash temperature change curve, and calculating the moisture absorption rate at the current moment based on the mash moisture change curve; extracting the standard temperature rise rate and standard moisture absorption rate from the preset process target database, and comparing the current temperature rise rate with the preset process target database. The temperature deviation value is calculated by subtracting the standard temperature rise rate from the standard water absorption rate. The water absorption rate at the current moment is then subtracted from the standard water absorption rate to calculate the water deviation value. The temperature deviation value and the water deviation value are input into a preset feedback control algorithm. The feedback control algorithm performs calculations based on preset proportional coefficients, integral time, and derivative time to generate a water addition rate adjustment. This adjustment is then superimposed on the current water addition rate to obtain the adjusted water addition rate. The adjusted water addition rate is updated as the new current water addition rate, and a rate adjustment command is generated based on the adjusted water addition rate and sent to the water adding device.

[0050] The system transforms real-time acquired discrete data points into process curves with continuous trends. Specifically, the control system maintains two time-ordered data queues (or buffers): one stores data pairs (acquisition timestamp, mash temperature value), and the other stores data pairs (acquisition timestamp, mash moisture value). Whenever new sensor data arrives, it is appended to the end of the corresponding queue. In this way, the system constructs and maintains the mash temperature and moisture change curves in real time. To improve the accuracy of subsequent calculations, the system can also apply filtering algorithms to these raw data, such as using a moving average filter to smooth the most recent N data points, to eliminate potential sensor noise and instantaneous fluctuations.

[0051] Furthermore, the system calculates core indicators reflecting the current process dynamics based on the constructed process curves. In this embodiment, the system needs to calculate the instantaneous rate of change at the current moment. One specific implementation scheme is to use the first-order difference method, that is, to subtract the data value of the previous moment from the latest data value, and then divide by the time interval (Δt) between the two data points, thereby approximating the current temperature rise rate and moisture absorption rate. As a better scheme, in order to obtain more stable and interference-resistant rate values, the system can take the most recent M data points (e.g., M=5), perform linear fitting on these points using the least squares method, and use the slope of the obtained straight line as a more accurate rate estimate for the current moment.

[0052] Furthermore, the system quantitatively compares the actual process progress with the preset ideal target to determine the deviation. The system has a pre-set process target database, which stores standard process curve parameters derived from expert experience or historical data analysis. Specifically, this database can be a lookup table that defines the standard temperature rise rate and standard moisture absorption rate corresponding to different stages of the water addition process (e.g., indexed by the amount of water added or the time elapsed). The control system retrieves the corresponding standard rate value from this database based on the current actual process progress (e.g., the current cumulative water addition is 30% of the total). Subsequently, the system performs two simple difference calculations: subtracting the calculated current temperature rise rate from the standard temperature rise rate to obtain the temperature deviation value; and subtracting the current moisture absorption rate from the standard moisture absorption rate to obtain the moisture deviation value.

[0053] Furthermore, the system utilizes a classic feedback control algorithm to convert the aforementioned deviation values ​​into specific equipment adjustment commands. In this embodiment, the system employs a multivariable PID (Proportional-Integral-Derivative) controller. This controller receives temperature deviation and moisture deviation values ​​as inputs. It calculates these two deviation values ​​based on pre-set proportional coefficients (P), integral time (I), and derivative time (D), comprehensively considering the current deviation magnitude (P), the cumulative historical deviation (I), and the future deviation trend (D), ultimately calculating and generating a single water addition rate adjustment. This adjustment can be a positive value (indicating an increase in rate) or a negative value (indicating a decrease in rate). The system then superimposes this adjustment with the current water addition rate to obtain an adjusted water addition rate and immediately updates it to the new command rate. Finally, this updated rate value is encoded into a new control command and sent to the PLC and frequency converter of the water addition equipment, thereby completing a closed-loop feedback adjustment and ensuring that the entire water addition process accurately follows the preset optimal process path.

[0054] Optionally, the step of inputting the temperature deviation value and the moisture deviation value into a preset feedback control algorithm, and generating a water addition rate adjustment amount by the feedback control algorithm according to a preset proportional coefficient, integral time, and derivative time, includes: weighting and summing the temperature deviation value and the moisture deviation value according to a preset weighting coefficient to obtain a comprehensive deviation value; multiplying the comprehensive deviation value by the preset proportional coefficient to obtain a proportional adjustment component; performing a time integral operation on the comprehensive deviation value, and dividing the integral operation result by the preset integral time to obtain an integral adjustment component; performing a time derivative operation on the comprehensive deviation value, and multiplying the derivative operation result by the preset derivative time to obtain a derivative adjustment component; summing the proportional adjustment component, the integral adjustment component, and the derivative adjustment component to obtain a preliminary adjustment amount; and performing amplitude limiting processing on the preliminary adjustment amount to limit the preliminary adjustment amount within a preset adjustment range to generate the water addition rate adjustment amount.

[0055] Specifically, the system assigns a preset weighting coefficient to both the temperature deviation and moisture deviation values. For example, the weighting coefficient W_m for moisture deviation can be set to 0.6, and the weighting coefficient W_t for temperature deviation can be set to 0.4. These two weighting coefficients reflect the importance of the corresponding parameters to the final product quality at the current stage of the process. Subsequently, the system performs a weighted summation calculation: Overall deviation value = Moisture deviation value × W_m + Temperature deviation value × W_t. In this way, the system integrates two independent control variables into a single, comprehensive deviation value that fully reflects the degree of deviation from the current process state, providing a unified and clear input for subsequent PID calculations. These weighting coefficients are configurable and can be optimized and adjusted according to different mash types or production seasons.

[0056] Furthermore, the system calculates the three core control components of the PID controller in parallel based on this comprehensive deviation value. First is the proportional control component (P), which aims to provide an immediate, proportional response to the deviation at the current moment. The system multiplies the comprehensive deviation value by a preset proportional coefficient (Kp) to obtain the proportional control component. The function of this component is that the larger the deviation, the stronger the control effect, thereby quickly pulling the process state back to near the target value. Second is the integral control component (I), whose purpose is to eliminate any potential steady-state errors in the system, namely those small but persistent deviations that the proportional element cannot completely eliminate. The system performs time integration on the comprehensive deviation value within one control cycle (usually summed in digital implementation), and then divides the sum by a preset integral time (Ti) to obtain the integral control component. This component accumulates the deviation over time; as long as the deviation exists, the integral action will continue to strengthen until the deviation is completely eliminated. Finally, there is the derivative control component (D), which is used to predict the trend of deviation changes, playing a role in early suppression and stabilization of the system. The system performs a time derivative operation on the overall deviation value (in digital implementation, this is usually calculated as the difference between the current deviation and the previous deviation), and then multiplies the rate of change by a preset derivative time (Td) to obtain the derivative adjustment component. When the deviation changes too quickly, this component generates a reverse suppression effect, thereby preventing the system from experiencing excessive overshoot and oscillation.

[0057] Furthermore, after calculating the three independent adjustment components, the system linearly superimposes them to form a comprehensive adjustment command. Specifically, the control system performs a summation operation: Initial adjustment = proportional adjustment component + integral adjustment component + derivative adjustment component. This initial adjustment theoretically reflects the total adjustment range of the water addition rate required to correct the current comprehensive deviation. This summation operation integrates the response to the current deviation, the correction of historical deviations, and the prediction of future deviations, embodying the essence of the PID control algorithm and enabling fast, accurate, and stable control.

[0058] Furthermore, to ensure safe equipment operation and process stability, the system performs a crucial protective step—amplitude limiting—before outputting the final control command. An unrestricted initial adjustment might, under extreme conditions (such as system startup or significant external disturbances), calculate a value that is either extremely large or extremely small. If directly applied to pumps or valves, this could cause equipment shocks or drastic process fluctuations. Therefore, the system presets a reasonable adjustment range, such as [-5 liters / minute, +5 liters / minute]. The system checks the calculated initial adjustment; if its value exceeds this range, it forcibly limits it to the range's boundaries (e.g., if the calculated result is +7, the final output is +5; if the calculated result is -8, the final output is -5). The value after this amplitude limiting process is the final, safe, and effective water addition rate adjustment, which will be used to update the current water addition rate.

[0059] Optionally, the method further includes: measuring the moisture content of the mash after water addition using a preset moisture detection device to obtain the measured moisture value of the mash; extracting the preset moisture standard corresponding to the fermentation pit to be treated from a preset process standard database, calculating the difference between the measured moisture value of the mash and the preset moisture standard to obtain a moisture control deviation value; determining the prediction accuracy of the intelligent water control model based on the magnitude and direction of the moisture control deviation value, and calculating the model weight adjustment coefficient and the model... The model is modified by adjusting the bias coefficient; the measured values ​​of mash moisture, ambient temperature and humidity, target water addition, and measured moisture content of mash entering the fermentation tank are combined into training sample data; the training sample data and the model modification parameters are input into the intelligent water control model; the intelligent water control model adjusts the weight parameters and bias parameters within the model according to the model modification parameters, and updates the network parameters of the intelligent water control model through a preset backpropagation algorithm to obtain the target intelligent water control model.

[0060] As a high-level embodiment of the method described in this application, to ensure the long-term accuracy of the intelligent water control model and its adaptability to changes in the production environment, this solution also introduces a closed-loop model self-correction and iterative optimization mechanism based on actual production results. The first step of this mechanism is to accurately quantify and evaluate the control effect after each water addition operation. Specifically, when a batch of mash has completed all processes such as water addition and grain moistening and is ready to enter the fermentation pit, the system will use a preset moisture detection device to finally determine its final moisture content, thereby obtaining the measured value of the moisture content of the mash entering the pit. This detection device can be a near-infrared online moisture analyzer installed at the end of the conveyor belt, or an automatic sampling arm combined with a rapid moisture analyzer set at the pit inlet.

[0061] Furthermore, the system compares the measured result with the process standard to generate a deviation signal. The system extracts the corresponding preset moisture standard from the preset process standard database, based on the current fermentation pit number, shift, or batch information. This standard value is the ideal moisture content to ensure optimal fermentation of the mash in this fermentation pit. Next, the system performs a simple difference calculation, subtracting the preset moisture standard from the measured moisture content of the mash entering the pit to obtain the moisture control deviation value. This deviation value visually quantifies the gap between the final effect of this intelligent control and the ideal target; its sign and magnitude precisely reflect whether the prediction is too high or too low, and the degree of deviation.

[0062] Furthermore, based on the aforementioned quantified deviation, the system generates correction parameters to guide model adjustments. The system determines the accuracy of the model's prediction based on the absolute value of the moisture control deviation; a smaller deviation indicates higher accuracy. Then, the system calculates the model weight adjustment coefficient and model bias adjustment coefficient using a preset mapping function (e.g., a function proportional to the deviation value) and combines them into model correction parameters. One specific implementation is that when the deviation is positive (indicating higher final moisture content), the generated correction parameters tend to reduce the model's output in the next prediction; conversely, when the deviation is negative, they tend to increase the model's output. Simultaneously, the larger the absolute value of the deviation, the larger the absolute value of the generated correction coefficient, meaning a stronger correction to the model.

[0063] Furthermore, while generating the correction parameters, the system organizes and encapsulates the entire process data of this operation to form a new, high-quality training sample. Specifically, the system combines three input data points—the moisture content of the mash measured at the start of the operation, the ambient temperature and humidity measured during the process, and the target water addition initially predicted by the model—with the final measured moisture content of the mash entering the fermentation tank, forming a complete training sample. In this sample, the first three constitute the model's input feature vector, while the latter constitutes the true label for this prediction. It comprehensively records the entire process of what decisions the model made under what initial conditions, and what actual results were ultimately achieved.

[0064] Furthermore, the system utilizes the newly generated training samples and correction parameters to perform online or offline training on the intelligent water control model itself. In this embodiment, the system inputs the aforementioned training sample data and model correction parameters into the current intelligent water control model. After receiving this data, the model first uses the model correction parameters to perform a guided fine-tuning of the weights and biases of specific layers of the internal neural network. Subsequently, based on the training samples, the model initiates a backpropagation algorithm. In this algorithm, the model calculates the loss between its prediction based on the input features and the true label, and then, based on this loss value, calculates the gradient layer by layer along the backpropagation path of the network, and uses this gradient to update all weights and bias parameters in the network. This process is equivalent to the model learning from its successes or failures, enabling the updated target intelligent water control model to make more accurate predictions when facing similar working conditions in the future, thus achieving continuous evolution of the model.

[0065] Optionally, the step of determining the prediction accuracy of the intelligent water control model based on the magnitude and direction of the moisture control deviation value, calculating the model weight adjustment coefficient and model bias adjustment coefficient based on the prediction accuracy, and generating the model correction parameters includes: calculating the absolute value of the moisture control deviation value to obtain the deviation amplitude value; comparing the deviation amplitude value with a preset set of deviation level thresholds; determining the deviation level corresponding to the moisture control deviation value based on the comparison result; determining the direction of water addition control based on the sign of the moisture control deviation value; and determining the direction of water addition control when the moisture control deviation value is positive. If the water level is insufficient, excessive water addition is determined when the water control deviation value is negative. Based on the deviation level, the corresponding prediction accuracy score is retrieved from the preset accuracy mapping table, and a learning rate adjustment factor is calculated based on the prediction accuracy score. The learning rate adjustment factor is multiplied by the preset benchmark learning rate to obtain the current learning rate. The model weight adjustment coefficient and the model bias adjustment coefficient are calculated based on the current learning rate and the water addition control direction. The model weight adjustment coefficient, the model bias adjustment coefficient, and the current learning rate are encapsulated into a parameter data structure to generate the model correction parameters.

[0066] Specifically, the system first calculates the absolute value of the deviation to obtain a non-directional deviation magnitude value. The system internally pre-defines a set of deviation level thresholds, for example, {0.5%, 1.5%, 3.0%}. The system sequentially compares the deviation magnitude value with these thresholds to determine its deviation level. For example, a deviation magnitude of 0.3% belongs to Level 1 (minor deviation); 1.2% to Level 2 (moderate deviation); 2.5% to Level 3 (significant deviation); and exceeding 3.0% to Level 4 (serious deviation). Through this classification, the system can qualitatively assess the model's prediction error, laying the foundation for subsequent differential adjustments.

[0067] Furthermore, the system analyzes the sign of the moisture control deviation value to determine the specific direction of the model prediction error. In this embodiment, the moisture control deviation value is calculated by subtracting the preset moisture standard from the measured moisture content of the mash entering the fermentation tank. Therefore, when the deviation value is positive, it indicates that the final moisture content of the mash is higher than the standard value, meaning that the actual amount of water added by the system during the water addition process is excessive relative to the mash's actual water absorption capacity; conversely, when the deviation value is negative, it indicates that the final moisture content is lower than the standard value, meaning that the actual amount of water added is insufficient. The system records this water addition control direction (for example, -1 for insufficient and +1 for excessive), and this direction information will directly guide the direction of subsequent model parameter adjustments.

[0068] Furthermore, the system dynamically calculates the core parameter for this model optimization—the learning rate—based on the level and direction of the deviation. Internally, the system maintains a pre-defined accuracy mapping table, establishing a mapping relationship from deviation levels to prediction accuracy scores. For example, level one corresponds to a score of 0.95, level two to 0.7, level three to 0.4, and level four to 0.1. Higher scores indicate that the model's predictions are closer to perfection. The system calculates the learning rate adjustment factor based on the score using a pre-defined function (e.g., learning rate adjustment factor = 1 - prediction accuracy score). Thus, the larger the deviation and the lower the score, the larger the adjustment factor. Subsequently, the system multiplies this adjustment factor by a pre-defined baseline learning rate (e.g., 0.01) to obtain the current learning rate specific to this model iteration. This mechanism achieves adaptive learning rate adjustment: when the model performs well, fine-tuning is performed with smaller step sizes; when the model exhibits significant deviation, rapid correction is performed with larger step sizes.

[0069] Furthermore, after determining the learning step size (current learning rate) and direction (water addition control direction), the system begins to calculate the specific model parameter adjustments. One specific implementation is that the system multiplies the water addition control direction (valued at +1 or -1) by the current learning rate. This result defines the base gradient direction and magnitude for this model parameter update. The system uses this result as the base values ​​for the model weight adjustment coefficient and the model bias adjustment coefficient, respectively. In more complex implementations, this coefficient is also combined with the gradients calculated in the backpropagation algorithm to determine the precise update values ​​for each specific weight and bias. The core idea is that when excessive water addition is detected, the adjustment coefficient is positive, and the model parameters will be updated in the direction of reducing the predicted water addition value; conversely, the opposite is also true.

[0070] Furthermore, the system integrates all calculated adjustment-related information into a structured data object, namely the final model correction parameters. Specifically, the system creates a parameter data structure (such as a JSON object or class instance) containing at least three key fields: model weight adjustment coefficients, model bias adjustment coefficients, and the current learning rate. This information is encapsulated together to form a complete, self-contained correction instruction package. This model correction parameter object is then passed to the model's training module, which parses the data and uses it to precisely execute backpropagation and parameter update processes, thereby completing a precise, efficient, and adaptive iterative model optimization.

[0071] It should be noted that the above embodiments of the apparatus are only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the apparatus and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be found in the method embodiments, which will not be repeated here.

[0072] This embodiment also discloses an electronic device, as shown in the reference. Figure 3 The electronic device may include: at least one processor 301, at least one communication bus 302, user interface 303, network interface 304, and at least one memory 305.

[0073] The communication bus 302 is used to enable communication between these components.

[0074] The user interface 303 may include a display screen and a camera. Optionally, the user interface 303 may also include a standard wired interface and a wireless interface.

[0075] The network interface 304 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0076] The processor 301 may include one or more processing cores. The processor 301 connects to various parts of the server using various interfaces and lines, and performs various server functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in memory 305, and by calling data stored in memory 305. Optionally, the processor 301 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 301 may integrate one or a combination of several of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content required for display; and the modem handles wireless communication. It is understood that the modem may also not be integrated into the processor 301 and may be implemented as a separate chip.

[0077] The memory 305 may include random access memory (RAM) or read-only memory. Optionally, the memory 305 may include a non-transitory computer-readable storage medium. The memory 305 can be used to store instructions, programs, code, code sets, or instruction sets. The memory 305 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory 305 may also be at least one storage device located remotely from the aforementioned processor 301. Figure 3 As shown, the memory 305, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and an application program for a method of controlling the amount of water used in the brewing process of baijiu (Chinese liquor).

[0078] exist Figure 3 In the electronic device shown, the user interface 303 is mainly used to provide an input interface for the user and to obtain the user input data; while the processor 301 can be used to call the application program stored in the memory 305, which is a method for controlling the amount of water in the process of brewing liquor. When executed by one or more processors 301, the electronic device executes one or more methods as described in the above embodiments.

[0079] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0080] In the several embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the shown or discussed mutual couplings or direct couplings or communication connections may be through some service interfaces; indirect couplings or communication connections between apparatuses or units may be electrical or other forms.

[0081] The foregoing description is merely an exemplary embodiment of this disclosure and should not be construed as limiting the scope of this disclosure. Any equivalent changes and modifications made in accordance with the teachings of this disclosure shall still fall within the scope of this disclosure. Other embodiments of this disclosure will be readily apparent to those skilled in the art upon consideration of the disclosure in this specification. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not described in this disclosure. The specification and embodiments are to be considered exemplary only, and the scope and spirit of this disclosure are defined by the claims.

Claims

1. A method for controlling the amount of water added during the brewing process of Baijiu (Chinese liquor), characterized in that, The method includes: Collect historical production data of pits for a preset number of production cycles, clean and structure the historical production data of the pits, and generate a water quality process knowledge base. The moisture state characteristics of the mash and the parameters of the water-measuring process are extracted from the water-measuring process knowledge base. A correlation matrix is ​​obtained by performing a correlation analysis on the moisture state characteristics of the mash and the water-measuring process parameters. Target process factors are selected based on the correlation matrix, and a water intelligent control model is trained and generated based on the target process factors. The moisture content of the mash and the ambient temperature and humidity of the fermentation pit to be treated are obtained. The moisture content of the mash and the ambient temperature and humidity are input into the intelligent water addition control model to calculate the target water addition amount and the initial water addition rate. A water addition control command is generated based on the target water addition amount and the initial water addition rate and sent to the water addition equipment to execute the water addition operation. During the water addition operation, the temperature change data and moisture change data of the mash are collected in real time. Based on the temperature change data and moisture change data of the mash, the deviation value between the current process state and the preset process target is calculated, and the current water addition rate of the water addition equipment is adjusted according to the deviation value until the water addition operation is completed.

2. The method according to claim 1, characterized in that, The correlation matrix obtained by performing a correlation analysis between the moisture state characteristics of the fermented mash and the water mixing process parameters includes: The moisture state characteristics of the fermented mash and the water mixing process parameters are normalized to obtain a standardized feature dataset; Based on the standardized feature dataset, the Pearson correlation coefficient between the moisture state characteristics of each mash and the water mixing process parameters is calculated, and a correlation coefficient matrix is ​​generated. Obtain the number of samples in the standardized feature dataset and calculate the degree of freedom of the samples. Determine the significance probability value corresponding to each correlation coefficient based on the correlation coefficient matrix, the degree of freedom of the samples, and the preset probability distribution model. Select the correlation coefficients whose significance probability value is less than the preset significance level threshold as effective correlation coefficients. Extract the target mash moisture state feature identifier and the target water process parameter identifier corresponding to the effective correlation coefficient. The target mash moisture state feature identifiers are sorted according to a preset sorting rule and mapped sequentially to matrix row indices. The target water process parameter identifiers are sorted according to a preset sorting rule and mapped sequentially to matrix column indices. A two-dimensional matrix index structure is constructed based on the matrix row indices and the matrix column indices. Based on the target mash moisture state characteristic identifier and the target volume water process parameter identifier, the corresponding row index value and column index value are found in the two-dimensional matrix index structure. The filling coordinates are determined based on the row index value and column index value, and the effective correlation coefficients are filled to the corresponding index positions in the two-dimensional matrix index structure according to the filling coordinates to generate the correlation matrix.

3. The method according to claim 1, characterized in that, The process of acquiring the moisture content of the mash and the ambient temperature and humidity of the fermentation pit to be treated, and inputting these values ​​into the intelligent water addition control model to calculate the target water addition volume and initial water addition rate includes: The moisture content data of the mash in the fermentation pit to be treated is collected by a preset moisture sensor to obtain the moisture detection value of the mash. Temperature and humidity data of the environment where the pit to be treated is located are collected by a preset temperature and humidity sensor to obtain the environmental temperature and humidity detection values. The moisture content of the fermented mash and the ambient temperature and humidity are normalized to map them to a preset range, thereby generating normalized input data. The normalized input data is combined into an input feature vector according to a preset feature arrangement order, and the input feature vector is input into the intelligent control model for measuring water. The input feature vector is calculated by forward propagation using the intelligent water control model, and the target water addition amount and the initial water addition rate are output through the output layer of the intelligent water control model.

4. The method according to claim 1, characterized in that, The step of calculating the deviation between the current process state and the preset process target based on the temperature change data and moisture change data of the mash, and adjusting the current water addition rate of the water addition equipment according to the deviation value, includes: The temperature change data and moisture change data of the mash are arranged in time series to construct temperature change curves and moisture change curves of the mash. The current temperature rise rate is calculated based on the temperature change curve of the mash, and the current moisture absorption rate is calculated based on the moisture change curve of the mash. The standard temperature rise rate and standard moisture absorption rate are extracted from the preset process target database. The temperature deviation value is calculated by the difference between the current temperature rise rate and the standard temperature rise rate. The moisture deviation value is calculated by the difference between the current moisture absorption rate and the standard moisture absorption rate. The temperature deviation value and the moisture deviation value are input into a preset feedback control algorithm. The feedback control algorithm performs calculations based on a preset proportional coefficient, integral time, and derivative time to generate a water addition rate adjustment amount. The adjusted water filling rate is superimposed with the current water filling rate to obtain the adjusted water filling rate. The adjusted water filling rate is then updated to the new current water filling rate, and a rate adjustment command is generated based on the adjusted water filling rate and sent to the water filling device.

5. The method according to claim 4, characterized in that, The step of inputting the temperature deviation value and the moisture deviation value into a preset feedback control algorithm, and generating a water addition rate adjustment amount by the feedback control algorithm based on a preset proportional coefficient, integral time, and derivative time, includes: The temperature deviation value and the moisture deviation value are weighted and summed according to a preset weighting coefficient to obtain a comprehensive deviation value; The comprehensive deviation value is multiplied by the preset proportional coefficient to obtain the proportional adjustment component; Perform time integration on the comprehensive deviation value, and divide the integration result by the preset integration time to obtain the integral adjustment component; Perform time differentiation on the comprehensive deviation value, and multiply the result of the differentiation operation by the preset differentiation time to obtain the differential adjustment component; The proportional adjustment component, the integral adjustment component, and the derivative adjustment component are summed to obtain the preliminary adjustment amount; The initial adjustment amount is subjected to amplitude limiting processing to restrict the initial adjustment amount within a preset adjustment range, thereby generating the water addition rate adjustment amount.

6. The method according to claim 1, characterized in that, The method further includes: The moisture content of the mash after water addition is measured using a pre-set moisture detection device, and the measured moisture value of the mash is obtained. Extract the preset moisture standard corresponding to the fermentation pit to be treated from the preset process standard database, calculate the difference between the measured moisture value of the mash entering the pit and the preset moisture standard, and obtain the moisture control deviation value. The prediction accuracy of the intelligent water control model is determined based on the magnitude and direction of the moisture control deviation value. The model weight adjustment coefficient and model bias adjustment coefficient are calculated based on the prediction accuracy to generate the model correction parameters. The measured values ​​of the mash moisture content, the measured values ​​of the ambient temperature and humidity, the target water addition amount, and the measured values ​​of the mash moisture content entering the fermentation tank are combined to form training sample data. The training sample data and the model correction parameters are then input into the intelligent water control model. The intelligent water control model adjusts the weight parameters and bias parameters within the model according to the model correction parameters, and updates the network parameters of the intelligent water control model through a preset backpropagation algorithm to obtain the target intelligent water control model.

7. The method according to claim 6, characterized in that, The process involves determining the prediction accuracy of the intelligent water control model based on the magnitude and direction of the moisture control deviation value, calculating the model weight adjustment coefficient and model bias adjustment coefficient based on the prediction accuracy, and generating the model correction parameters, including: The absolute value of the moisture control deviation value is calculated to obtain the deviation amplitude value. The deviation amplitude value is compared with a preset set of deviation level thresholds. Based on the comparison result, the deviation level corresponding to the moisture control deviation value is determined. The direction of water addition control is determined based on the positive or negative sign of the moisture control deviation value. When the moisture control deviation value is positive, it is determined that the water addition is insufficient; when the moisture control deviation value is negative, it is determined that the water addition is excessive. Based on the deviation level, the corresponding prediction accuracy score is retrieved from the preset accuracy mapping table, and the learning rate adjustment factor is calculated based on the prediction accuracy score. The learning rate adjustment factor is multiplied by the preset benchmark learning rate to obtain the current learning rate. The model weight adjustment coefficient and the model bias adjustment coefficient are calculated based on the current learning rate and the water addition control direction. The model weight adjustment coefficient, the model bias adjustment coefficient, and the current learning rate are encapsulated into a parameter data structure to generate the model correction parameters.

8. An electronic device, characterized in that, The device includes a processor, a memory, a user interface, and a network interface. The memory is used to store instructions. The user interface and the network interface are both used to communicate with other devices. The processor is used to execute the instructions stored in the memory to cause the electronic device to perform the method as described in any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed, perform the method as described in any one of claims 1-7.

10. A computer program product, characterized in that, When the computer program product is run on an electronic device, it causes the electronic device to perform the method as described in any one of claims 1-7.

Images